5 Questions to Ask Before Paying for Any AI Tool

Search "how to evaluate AI tools" and look at who wrote the top results. Salesforce. Amazon Business. Corporate IT blogs with a demo booking form in the sidebar. Every guide on choosing AI tools is written by someone selling one, and every checklist conveniently ends at their pricing page.

Here's my conflict of interest, stated upfront: I sell AI consulting services. And I still tell clients that most AI tools fail at least one of the five questions below. Usually more than one.

These are the questions I run through before any AI subscription touches my own credit card. They've killed far more purchases than they've approved. Steal them.

Everyone Writing These Guides Has a Tool to Sell You

When Salesforce publishes a guide on choosing the right AI tool for your business, one of the criteria is always "does it integrate with your existing CRM?" Guess whose CRM they have in mind. That's not evil, it's just marketing. But it means the advice you find when you search this topic is a sales funnel wearing a checklist costume.

There's a number floating around called the 10-20-70 rule: AI success is 10 percent algorithms, 20 percent technology and data, and 70 percent people and processes. Vendors love quoting it in their thought leadership. Then they sell you the 10 percent and wish you luck with the other 90.

I'm not neutral either. Businesses pay me to figure out whether they need AI help at all. But my incentive runs the other way: if I recommend a tool that quietly drains $200 a month and delivers nothing, I lose that client. Honest evaluation is literally my product. So here's the checklist I actually use.

Question Zero: Have You Tried This With What You Already Pay For?

Before the five questions, there's one almost nobody asks. You probably already pay $20 a month for ChatGPT or Claude. Have you actually tried doing this task with that subscription?

A frontier model with a decent prompt covers a shocking amount of the AI tool market. AI meeting summarizer? Paste the transcript. AI email writer? That's a chatbot with a template. AI content repurposer? "Turn this blog post into five LinkedIn posts" is one sentence. The AI tools that actually run my one-person agency are a short list, and the general-purpose subscriptions on it do more work than everything else combined.

If you haven't run this experiment, you're not evaluating a tool. You're shopping for a feeling.

Question 1: What Happens to Your Data?

Open the privacy policy and search the page for the word "train". That one search tells you more than the entire homepage. Is your data used to train their models? Where is it stored? Can you actually delete it, or does "delete" mean "hide from your dashboard"?

Then look for the sub-processor list. Many small AI tools are pipelines: your data goes to them, then to OpenAI or Anthropic, then to two analytics vendors you've never heard of. If you're in the EU like me, ask for a DPA (data processing agreement). A serious vendor sends it within a day. A hobby project with a nice landing page goes quiet.

My rule: if I can't answer "where does my data go" within five minutes of looking, that is my answer.

Question 2: If You Cancel Tomorrow, Can You Keep Working?

Before subscribing, find one thing in the documentation: the export function. Not the marketing page, the actual docs. What format do you get your data back in? And does the export include the prompts, workflows, and automations you'll spend months building, or just a CSV of your account settings?

If your setup lives in a proprietary format inside their app, you're not a customer. You're a hostage with a monthly payment plan. The price hike will come, and you'll pay it, because rebuilding six months of configuration costs more than the increase.

This is the same trap as every other SaaS, which is why I self-host half my stack. You don't need to self-host everything. You do need to know your exit cost before you enter.

Question 3: Is It a Wrapper?

A wrapper is a thin interface over the same API you could call yourself. The tell: the marketing says "powered by GPT-5", or refuses to name a model at all. There's no proprietary technology anywhere, just a text box, a hidden prompt, and a markup.

To be fair, a wrapper isn't automatically a scam. A good one saves real time because its workflow is shaped exactly like yours. But you should price it as convenience, not capability. If a tool charges $49 a month for what the underlying model does with $3 of API tokens, you're paying $46 for a nicer text box. Sometimes that's worth it. Usually it isn't.

The question to ask: what does this tool do that the model underneath doesn't? If the answer takes more than one sentence, it's a wrapper with good branding.

Question 4: Does the Free Tier Already Do Everything You Need?

Most AI tools gate volume, not capability. The paid tier gives you 500 generations instead of 25, priority support you'll never use, and team seats you don't have a team for. If your real usage is five times a month, the free tier is the product and the paid tier is a donation.

Watch out for credit-based pricing especially. Credits exist to make cost comparison impossible. If you can't calculate what one unit of your actual work costs, that's not an accident. That's the business model.

So run your real monthly usage against the free tier limits before paying. Numbers, not vibes.

Question 5: Would a Script Do It for Free?

A surprising share of "AI automation platforms" are a scheduled task plus one API call. Monitor a page for changes, summarize new entries, extract data from PDFs, rename files, draft a weekly report: that's 20 to 50 lines of code. And the AI subscription you already pay for will write those lines, even if you've never programmed anything.

I've replaced several paid tools this way. Not only because I enjoy building things (okay, also because I enjoy building things), but because a script I own has no monthly fee, sends no data to a third party, and can't disappear when the startup behind it pivots.

The honest caveat: scripts need occasional maintenance and someone who cares. If nobody in your business will touch it when it breaks, a paid tool with real support might genuinely be the better deal. That's a legitimate answer to this question, not a failure.

The 10-Minute Trial Protocol

Vendor demos are choreography. The sample data is chosen because it works. So here's how I test any AI tool before a credit card gets involved, and it takes ten minutes.

1. Bring your own real work. An actual document, an actual client email, an actual spreadsheet. Never the demo data.
2. Define "done" before you start. One sentence: "This tool earns its money if it produces X in Y minutes."
3. Set a timer for ten minutes and do the task in the tool.
4. Do the same task in ChatGPT or Claude with a plain prompt.
5. Compare the outputs side by side. Decide on results, not interface polish.

If a tool has no free trial and wants a card upfront, that's a signal too. Confident products let you test them.

And if you're evaluating on behalf of a boss who just watched an impressive AI demo, this protocol is your best defense. "I tested it with our real invoices, here's what happened" ends more bad purchases than any opinion ever will.

Most Tools Fail These Questions. That's Fine.

This is how to evaluate AI tools when nobody in the room is selling you one: most candidates fail at question zero, because the subscription you already have does the job. A few fail on data handling. Plenty fail the wrapper math. The tools that survive all five questions tend to be boring. They save hours every week on work you were already doing, and nobody posts breathless threads about them.

I'd love to tell you I follow my own checklist perfectly. I don't. I still buy tools out of curiosity, and some of them die quietly after a month. The difference is that I treat those as small experiments now, not as strategy.

I can't tell you which AI tool to buy without knowing your workflows. What I can offer: an honest look at where AI actually saves you time, including the possible answer "nowhere yet." Let's talk if that's the kind of advice you want.